The race to dominate artificial intelligence is hitting a hard wall: there simply isn't enough computing power to go around. Often called "compute," the processing capacity that fuels everything from chatbots to autonomous agents is now the most sought-after commodity in tech—and its scarcity is reshaping politics, markets, and corporate strategy.
"Modern AI systems don't work without compute," said Erich Grunewald, a senior researcher at the Institute for AI Policy and Strategy. "You need compute for almost everything, and more compute is almost always better."
Tech giants have responded with eye-popping investments. Last year, companies poured over $400 billion into AI infrastructure, and Bank of America projects that figure could hit $800 billion this year and exceed $1 trillion by 2027. But the physical limits of chips, servers, and data centers are colliding with soaring demand, especially as agentic AI—systems that act autonomously—takes off. A Goldman Sachs report from May warned that token consumption could surge 24-fold by 2030, reaching 120 quadrillion tokens a month.
The political landscape has shifted dramatically. Both President Trump and former President Biden issued executive orders to fast-track data center construction, but local opposition is mounting as communities push back against rising electricity bills and environmental concerns. Numerous projects have been stalled or blocked, complicating the bipartisan push for AI dominance. Meanwhile, supply chain bottlenecks persist. Chipmakers, accustomed to boom-and-bust cycles, are now grappling with what Janet Egan of the Center for a New American Security calls "real demand" that keeps climbing. "It takes many years to bring new facilities online," she noted.
Evidence of the crunch is everywhere. Nvidia's H100 chips, several generations old, are actually rising in price as demand outstrips supply. "Usually, older chips become cheaper over time, but demand has gone up so rapidly that even these are more expensive now than when they first launched," Grunewald observed. In a sign of the times, Google inked a deal with SpaceX for $920 million a month in computing capacity through 2029, while Anthropic is paying $1.25 billion monthly for access to SpaceX's Colossus 1 data center. Anthropic also recently raised usage limits on its Claude model after facing backlash for capping peak-hour access.
These constraints force companies into painful trade-offs. "If you're using compute to run experiments to improve your models, you have fewer resources for customers," Egan explained. "There's a direct trade-off between innovation and service." Despite fears of a total shortage, JPMorgan's Stephanie Aliaga argued that companies won't "run out" of compute, but "market conditions will increasingly reflect a supply-constrained environment, with pricing power accruing to providers of scarce inputs."
The stakes are so high that financial exchanges are jumping in. CME Group plans to launch compute futures contracts, with CEO Terry Duffy calling it "the new oil of the 21st century." As AI integrates deeper into daily life—from coding agents to autonomous systems—the battle for compute is only intensifying. For policymakers and industry leaders, the question is no longer whether to build, but how to navigate the political and economic bottlenecks that threaten to slow the AI revolution.
